A dynamic urban expansion model was established by integrating and taking advantages of cellular automata model and multi-agent simulation model. The model first simulated Shanghai urban expansion in 2005 and was tested, then it was used to project the expansion in 2010 and 2020. The data sources were mainly obtained from the Shanghai digital city data (including schools, hospitals, residential, commercial, industrial, parks and green spaces), digital elevation model, and social and economic data. The two parts, cellular automata model (CA) and multi-agent system (MAS), of the established model were integrated in the ArcGIS environment. The multi-agent system was used to overcome the deficiencies of cellular automata model. The influencing factors (such as social, economic, cultural and public policy) were included. The administrative districts of Shanghai were divided into urban center, suburb and periphery areas. The projection of urban land was based on the regression analysis of urban land, population and GDP from 1996 to 2009. The changes of other land use type were calculated and predicted by Markov chain model. The physical, social and transportation factors of urban systems were defined in the cellular automata model, including DEM, slope, and distances to watershed, airports, major road and railway. The conversion probability of urban land was estimated based on "bottom-up" spontaneous conversion. A binary logistic regression analysis was used to estimate the conversion probability of urban land. The behaviors of regional authorities and residents were defined in multi-agent system model, and their different roles in the process of land use conversion were defined according to their own characteristics. Each type of agents has different demands following their own rules of conduction, and each type of agent selected the locations of urban land conversion according to their preferences. The preferences of government agent were calculated by the distances to the locations of government. The preferences of resident agent were calculated by the distances to railway stations, parks, schools and hospitals. The final agreement was achieved by consultation. Urban expansion was determined by the behavior of agents, the interaction between different agents and between agent and environment. Multi-agent system model impacted on the driving factors of urban land conversion. Random factors in the final urban land conversion probability were taken into account at the end. According to the model, the Shanghai urban land in 2005 was simulated and compared with the actual land use status. The Kappa coefficient is above average to 0.75, indicating the model has a high credibility. Finally the land use of Shanghai in 2010 and 2020 were predicted, and the results showed that the urban land of Shanghai will expand eastward and southward in 2020. The findings and forecasting would be useful for city planning and decision making.